create_dl_transform_pipeline🔗
Short description🔗
create_dl_transform_pipeline — Create a pipeline of transforms for the training of a deep learning model.
Signature🔗
create_dl_transform_pipeline( dl_transform Transforms, attribute.name GenParamName, attribute.value GenParamValue, out dl_transform_pipeline TransformPipeline )
Description🔗
The operator create_dl_transform_pipeline creates a pipeline of
transforms Transforms for transforming a batch of samples
prior to their use in train_dl_model_batch and apply_dl_model.
Transforms may include adaptation of image dimensions or gray value range
as well as augmentation methods.
The execution order of the transforms is defined by their order in the
tuple Transforms. The execution is performed with the operator
transform_dl_sample_batch.
Note: The output image dimensions of the pipeline depend on the applied transforms, their order, and their parameters. If consistent output image dimensions are required (e.g., for training and inference pipelines), you can add a resize transform after the last transform that may change the image dimensions (e.g., random crop). If input dimensions are preserved in a transform (e.g., when using a resize transform configured accordingly), the output dimensions may vary between images.
The following generic parameters GenParamName and the corresponding
values GenParamValue are supported:
-
'fill_value': Value used to fill the padded regions if required by the transforms.
Restriction: Only values of type integer or float are allowed.
Default: 0.0.
-
'bbox_min_side_length': Minimum allowed side length of a transformed box instance. Smaller box instances are not transferred to output.
Default: 2
-
'instance_min_visibility': Minimum allowed visibility fraction of a transformed instance. Instances with lower visibility are not transferred to output.
Default: 0.5
-
'ignore_direction': Determines whether directions of oriented bounding boxes shall be used ('false') or ignored ('true'). Only for transformation of samples for detection models of 'instance_type' = 'rectangle2'.
List of values: 'true', 'false'
Default: 'false'
Note that additionally 'class_ids_no_orientation' can be set
using set_dl_transform_pipeline_param.
Execution information🔗
Execution information
-
Multithreading type: reentrant (runs in parallel with non-exclusive operators).
-
Multithreading scope: global (may be called from any thread).
-
Automatically parallelized on internal data level.
This operator returns a handle. Note that the state of an instance of this handle type may be changed by specific operators even though the handle is used as an input parameter by those operators.
Parameters🔗
Transforms (input_control) dl_transform(-array) → (handle)
Tuple with the transform handles
GenParamName (input_control) attribute.name-array → (string)
Names of the generic parameters that can be adjusted for the transform pipeline.
Default: []
List of values: 'bbox_min_side_length', 'fill_value', 'ignore_direction', 'instance_min_visibility'
GenParamValue (input_control) attribute.value-array → (real / integer)
Values of the generic parameters that can be adjusted for the transform pipeline.
Default: []
Suggested values: 0.0
TransformPipeline (output_control) dl_transform_pipeline → (handle)
Transform pipeline handle
Combinations with other operators🔗
Combinations
Possible predecessors
create_dl_transform_flip, create_dl_transform_normalize, create_dl_transform_random_crop, create_dl_transform_random_geometric, create_dl_transform_random_hsv, create_dl_transform_resize
Possible successors
get_dl_transform_pipeline_param, set_dl_transform_pipeline_param, transform_dl_sample_batch
Module🔗
Foundation